API

This page shows the public Python API for AutoNeuroNet.

Core types

• Var: Scalar value with gradient tracking and reverse-mode automatic differentiation.
• Matrix: 2D matrix of Var values with math, activation functions, gradients, and more.

Neural Network

• Layer (base class)
• Linear, ReLU, LeakyReLU, Sigmoid, Tanh, SiLU, ELU, Softmax
• NeuralNetwork

Optimizers

• GradientDescentOptimizer
• SGDOptimizer

Losses

• MSELoss
• MAELoss
• BCELoss
• CrossEntropyLoss
• CrossEntropyLossWithLogits

Utilities

• matmul(A, B)
• numpy_to_matrix(array, as_column=False)
• from_numpy(array, as_column=False) (alias)

Automatic Differentiation

Var

• Create a value: Var(1.0)
• Read/write value: v.val
• Read/write gradient: v.grad
• Backpropagation: v.setGrad(1.0); v.backward()

Matrix

• Create: Matrix(rows, cols)
• Index: M[i, j] or row slices M[i]
• Math: +, -, *, /, @, pow, sin, cos, tanh, relu, softmax, etc.

Neural Networks

NeuralNetwork

• Construct with a list of layers.
• Forward pass: model.forward(x)
• Save/restore weights: model.saveWeights(path), model.loadWeights(path)

Optimizers

• Create with learning_rate and a model.
• Call optimize() after backprop.
• Call resetGrad() between steps if needed.